(Quick Canvas exercise. Only available under the Gemini 2.0 Flash model. Took a couple of tries to set context, but allowed interactive questions to be appended to itts text and then used a side menu to select length of answer or bring up a prompt box, also prototyped diagram and offerred multiple windows for preview, code, and a console while logging everything in the conversation here.)
(This is fast. Waltzed through the questions. But users is still leading it on to a large degree and is going to remain crucial in identifying gaps. Can drag it through the other reasoning tests over time. In this case, the machine would be an observer as well. It does appear to now what other conversations it has had with the users, including the other models. Have to regression test to figure out if it is making the same logical errors, e.g. conflation. These all want to write and code. It boasts of larger context windows compared to other brands. 2.5 does not seem to know as much about itself as it does about 1.5. Using Imagen 3. Accompanying Gems are still at 2.0 Fllash as is Canvas. Friends are offered a 4-month free trial under Google One. 2 TB storage can be shared with up to 5 other people or upgraded to 5/10/20/30 TB by subscription..This is separate from Google Workspace or AI Studio API. It has an app. It can save custom info about the user as background for interactions. Others, such as Alibaba, are criticizing the data center expansion fad and whether these frequent rollouts will persist, but the issue is whether that is ideological and looking for other compatible solutions than maximizing debt leverage which is pure capitalism. Perhaps someone will distill the governments. Blame Pythagorus.)
(Various questions on how to review footnotes via other than a social network, blog, or website. Emphasis on new Google Gemini Canvas. Toward a Utility of Intelligence. Also writing off-digital in an A5 notebook, Moleskine this month. Wants to be hybrid, too, BuJo if not Deep Work. The bringup is like a deus in the box. All intuition, no indexing. Incidentally, Google Keep had the bookmarks for past trials and that was integrated since alongside calendar and gmail., although it does not import bookmark files such as HTML the way the browser does, but it can export using Google Takeout. Raindrop itself can do labeling or semantic search, but not document editing. AmpleNote has an AmpleCap extension to capture pages to notes, and any AI interaction is through a third-party platform API subscription.)
(Compare to pro/con, evaluation, and rebuttal. The method of studying foreign languages would also show variations in thinking styles. This particular Deep Research topic did not include Mao on contradiction. In practice Gemini tends to simplify or conflate terms like cybernetics and post-cybernetics, e.g. when discussing critiques. That may be an invitation to write up more sources so it can expand or expound upon the blanks. Empirical research could demand its own models. Or may be able to setup the browser or an app so the options include a menu of methods or chatbots that the research agent can circulate among in its pursuit. Then it likely runs into the need to simulate a computer that could indeed solve the problem. Or maybe from the machine side, they emulate a virtual person. Yet another merger of dynasties.)
Re: AAAI 2025 Future of AI Research. (Basically, current excitement, taken as hype, includes some things that are real but need a lot more. The next wave of participants has to determine what they want. The claim that AGI is the next milestone seems like a running punchline. Their task is to figure out what paths are feasible and for which purposes. Having said that, new capabilities enabled by the tech may also pop up and be looking for applications. Not quite theory or modeling, but classic tech. Of course, quantum could claim that Feynman was wrong, too, but somehow that persists. Action does not always depend upon understanding. A breakout from the Cave.)
New Gemini app features, available to try at no cost
(Tried running a shopping list through Gemini 2.0 for pricing from some stores and got mixed accuracy. It also had to step through one at a time. Then instructed it to use Google Search to verify all the items from a single store and it matched the store's site. Contrast Google AI Mode which balked at the same list.)
Gemini Deep Research prompt: What are the different types of problem sets from the AI perspective, including new capabilities such as embodiment, sensors, and agents?
(Research plan on 2.0 looks similar to 1.5. Initial claims from training were sequentially verified against different search sites. Growing number of references. Perhaps the idea would be to figure out a starting prompt that will then take it through skill acquisition to grand challenges, or exploit multiple agents in parallel, or cover sectors or countries, or integrate datasets, or find altogether new questions. Personalization could make it take on synthetic roles, not unlike early Eliza, but now headed for the C-Suite or whoever assures equity under ideology. Without triggering Hulk.)
Google Gemini App on X: "5⃣ Deep Research is even smarter, rolling out for everyone to try. Today, we're giving Gemini users the ability to try Deep Research, your personal AI research assistant, at no cost. 🧠✨ Gemini Deep Research has been upgraded to 2.0 Flash Thinking Experimental. With this https://t.co/gMga5dJUvo" / X
📁 Document upload is now available to all Gemini users. Upload multiple Google Docs, PDFs, and Word documents from Google Drive or your device. Get quick summaries, personalized feedback, and actionable insights—streamlining your workflows and saving you time. Tap the plus… pic.twitter.com/8FS49dGlbz— Google Gemini App (@GeminiApp) February 20, 2025